How this site is made

Why this page exists

Google's guidance on helpful content asks creators to be able to answer three questions about any use of automation: whether it is self-evident to visitors, whether there is background on how it was used, and whether there is an explanation of why it was useful. Those are fair questions and the honest answers are more interesting than a disclaimer.

There is also a version of this site that would not have been worth making. If the plan had been to generate a few hundred pages of plausible text about heating and hope some of it ranked, the right thing to do would have been to not build it. What makes this worth publishing is a dataset and a model that did not exist in this form before, and I want to be specific about which parts those are.

What is computed

This is the part I would defend hardest, and it is code rather than prose.

The data pipeline

A script in this project calls the U.S. Energy Information Administration public API and pulls five series: residential electricity price by state, residential natural gas price by state, residential propane and heating oil prices by PADD region, and heating degree day normals by census division. It averages the price series over the most recent complete December to February window, using identical months for every state, and writes a single data file.

Nothing in that file is typed in by hand. If a state is missing a month it is excluded rather than estimated, which is why the site covers 51 states and not fifty.

The model

A second file holds the physics: unit conversions, the HSPF2 to COP conversion, the harmonic combination for backup heat, and the breakeven relationship. I derived the breakeven from first principles rather than adopting a published rule of thumb, and checked it two ways that agree.

What follows from those two

Every price, ratio, verdict and cost figure on this site, in the calculator and in the writing, comes out of those two files. There is no second copy of the numbers, which is why the tool and the articles cannot contradict each other.

What is written, and how AI was used

The explanatory pages, the comparison guides and the field notes are prose. They were drafted with AI assistance, in a working session where I directed the research, chose the angles, and checked the output.

What that actually looked like

• I decided what to write about, after looking at what already ranks for these queries and where the existing results were thin.
• Research was done against primary sources. Where this site cites a federal field study, that report was read directly rather than summarised from someone else's coverage. PNNL-37127 is a 66 page PDF and the figures quoted here came out of it, not out of a press release about it.
• Drafting was AI-assisted. Structure, phrasing and first drafts.
• Every figure was checked against source before publication. That check is not decoration. It is what caught the errors below.

What the checking caught

Two things worth naming, because they are the argument for why the checking matters.

The first build of the data pipeline used the most recent monthly EIA gas price, which for gas resolved to a summer month. EIA's residential price is revenue divided by volume, and in summer the fixed customer charge is divided by almost no gas, inflating the apparent price by up to 114%. The site was answering a winter question with a summer number and it biased every verdict toward the heat pump. Georgia was overstated by 114%, and fixing it changed the answer in most of the country.

Separately, the site stated that roughly 40% of US homes do not heat with natural gas. The real figure is higher: gas is the main heating fuel in 47% of homes, so slightly more than half do not. That was a claim carried from memory rather than checked, and it was wrong.

Both are written up in full in the corrections log, with the old and new figures side by side.

The state pages, and why they look different now

There are 51 state pages, one per state with complete data. They are generated from the data file.

They were originally built as articles: around 900 words each, with a headline that branched on the verdict, an FAQ section, and Article and FAQPage structured data. That was a mistake and I have changed it.

They are now presented as what they are. Shorter, table led, explicitly labelled as data pages, marked up as datasets rather than articles, with no invented FAQ. The analysis they used to pad out with links out to the pages where the reasoning actually lives.

What this site does not have

I have not installed a heat pump, lived with one through a winter, or opened a January electricity bill in Vermont. That gap is real and I cannot close it by reading more carefully.

I have started closing the part of it that is reachable. I called three contractors and asked them the same question on the record, and I paid for a Manual J load calculation so that the site's least defensible assumption could be tested against a real document rather than asserted. Both are written up in asking three contractors the same question, including the household data I collected and then declined to use because it did not survive checking.

I am working on closing it. If you have a heat pump and a year of bills, or you are an installer willing to talk about balance points and how you size systems, I would genuinely like to hear from you at banerjeeashutosh01@gmail.com. Real numbers from real houses would improve this site more than anything else I could do to it.

Who is behind it

Ashutosh Banerjee, who goes by Ash. B.Tech in Mechanical Engineering from Sardar Vallabhbhai National Institute of Technology, Surat, 2024. The certificate is on the about page because a claimed credential is worth less than a shown one.

The thermodynamics here is undergraduate mechanical engineering. Coefficient of performance, seasonal efficiency and fuel heat content are standard material, which is exactly why the assumptions can be written down and checked rather than asserted.

I sell no heating equipment, take no money from manufacturers, installers or utilities, and the site currently carries no advertising and earns nothing. If that changes it will be disclosed here and on the editorial standards page before any money changes hands.

How this was checked

This page is a description of process rather than a claim about the world, so there is less to check than usual. The two corrections it describes are documented in full in the corrections log with their figures, and the pipeline it describes is the one that produced the data file currently powering the site, dated 2025-12, 2026-01, 2026-02.

What this page does not cover

This page does not cover the model itself, which is in the methodology, or the specific sources behind individual claims, which are cited on the pages that make them.

References

1. U.S. Energy Information Administration. Electricity Data Browser, retail sales and average price by state and sector, monthly. Series accessed through the EIA Open Data API v2.
2. U.S. Energy Information Administration. Natural Gas Prices, residential price by state, monthly, dollars per thousand cubic feet. Series accessed through the EIA Open Data API v2.
3. U.S. Energy Information Administration. Weekly Heating Oil and Propane Prices, residential, by region.
4. U.S. Energy Information Administration. Short-Term Energy Outlook, heating degree days by census division, prior ten year average series. Accessed through the EIA Open Data API v2.
5. U.S. Energy Information Administration (2025). Electricity use is becoming more common for residential heating. Today in Energy, citing U.S. Census Bureau American Community Survey and the Residential Energy Consumption Survey.